Neural-symbolic computing aims to integrate the two most fundamental cognitive abilities according to Leslie Valiant: the ability to learn from experience, and the ability to reason from what has been learned.
Here, knowledge is represented in symbolic form, while learning is computed through neural networks and reasoning conducted on them via logical reasoning systems. This provides a framework capable of bridging lower-level information processing (for perception and pattern recognition) and higher-level abstract knowledge (for reasoning and explanation).
Currently, we have two projects in the space of Neural Symbolic Computing:
There are potentially many applications where human knowledge is encoded as symbolic formula and applied to some sort of learning. Consider the case where a medical image needs to be classified to detect a disease. In this setting, we can use symbolic reasoning to:
Leveraging existing medical knowledge for improved model learning.
Explaining predictions for human interpretability that satisfy compliance and reassure medical personnel.